JPH0337230B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a magnetic ink character recognition method, which is used in financial institutions to process checks, bills, and the like.

Magnetic ink character readers (MICR) have been introduced in banks for some time now, and the MICR characters printed on checks and bills are automatically read, sorted, and tabulated. E13B as a character for MICR
Type characters (JIS C6251) are often used,
Various reading methods have been proposed. As an example, JP-A-56-145467 discloses that an analog signal read by a magnetic head is converted into a digital signal of "1" or "0" corresponding to the presence or absence of a peak, and this is converted into a standard signal for each character. A method of comparing with pattern information is shown. Although this method simplifies the device configuration, the height and position of the peak may vary due to differences in the print quality of MICR characters, differences in the relative position of the magnetic heads, or aging of the check or device. The disadvantage is that it is easy to misrecognize characters.

In many cases, the information read by the reading device becomes input information directly into the computer, and even if multiple check points are provided, errors in recognizing letters and numbers will hinder the smooth operation of banking operations. In some cases, it may affect the credibility of customers.

The present invention was made in view of the above circumstances, and
It is an object of the present invention to provide a character recognition method that does not easily cause character recognition errors even if there are slight variations in reading with a magnetic head due to usage conditions of a check or the like or environmental conditions.

The present invention A/D converts an analog signal at each peak position, detects the peak value for each mesh in the horizontal direction of the character, converts it into character data, normalizes the character data, and then This magnetic ink character recognition method is characterized by identifying characters by comparing them with standard pattern data.

Hereinafter, the present invention will be explained using examples and drawings.

In Figure 1, a permanent magnet 1 is placed on the upstream side in the direction of travel of check C, and a reading head 2 is placed on the downstream side.
It reads the E13B type characters electromagnetically and outputs an analog signal. The amplifier 3 amplifies this analog signal to an appropriate level, and in order to remove unnecessary signals due to noise and printing defects, the amplifier 3 amplifies the advance speed of check C and the length of one mesh in the horizontal direction of the E13B type character. It has a built-in low-pass filter with a cutoff frequency determined by the The output of the amplifier 3 is input to the peak detector 4, which detects the positive and negative peak positions and outputs a peak position signal. At that timing, the A/D converter 5 is started to calculate the amplitude (peak value) at each peak position. ) is A/D converted. Such a peak detector 4, for example, creates positive-side and negative-side pulse signals using appropriate threshold values from each of an input analog signal and its differential signal, and generates pulse signals on the same side. It is sufficient to use the trailing edge of the output pulse signal obtained by performing the logical product of . A/D converter 5
calculates each peak value of the input analog signal by A/
At the same time, the data is converted into a digital signal in the form of a sign and an absolute value, and is sequentially outputted to the next character cutting section 6.

The character cutting section 6 is for dividing the continuously input data of peak positions and their peak values into mesh units and character units, and constructing character data of one character. Referring to FIG. 2, when reading the character of stroke 8, the analog signal S of the output of amplifier 3 will be explained.
1 and the pulse signal S2 for each peak position output from the peak detector 4, the A/D converter 5 outputs a signal S3 obtained by A/D converting the amplitude of each peak value. 6 monitors the input signal S3 from the A/D converter 5 for a time T/2 corresponding to half a mesh from that time point when the positive peak at the beginning of a character is input. Then, data with the largest amplitude among the signals S3 input during this period is outputted to the memory section 7 and the maximum value extraction section 8 as first mesh data. Next, the input signal S3 is monitored for a time T corresponding to one mesh, and the data with the maximum amplitude during that time is outputted to the memory section 7 and the maximum value extraction section 8 as second mesh data. Input signal S
If there is no 3, output "0". After repeating this up to the 8th mesh, the time for 1 mesh is T.
, and then detects the positive peak at the beginning of the next character. The character cutting section 6 is
In detecting the peak at the beginning of a character, a relatively large constant positive value threshold is set.
Only when the peak value exceeds a threshold value is it determined that the peak is the first peak of a character, and subsequent predetermined operations are performed.

The memory section 7 receives data from the character cutting section 6 and sequentially stores it, and uses a FIFO element or a memory element addressed by the character cutting section 6. The maximum value extraction section 8 is for extracting the mesh having the maximum absolute value of amplitude from the data for one character from the character cutting section 6. It is set at the first mesh, and the data is extracted for each mesh. The maximum value in one character, its sign, and mesh number are retained.

The normalization control unit 9 reads data from the memory unit 7 and the maximum value extraction unit 8 and sends it to the division unit 10.
The dividing unit 10 performs normalization by dividing the absolute value of the amplitude of each mesh by the absolute value of the maximum amplitude within that character. The normalized character data is compared with the standard pattern of each character stored in the pattern section 12 in the identification section 11 to perform identification. Identification part 11
Now, perform the following calculation.

Ei= ₈ 〓 ^j=1 |Xj−Pij| (1) Here, Ei is the error function value for the character of stroke i, Xj is the normalized input data of the j-th mesh, and Pij is the j-th character of stroke i. This is the standard pattern data for the mesh. The error function value Ei is determined for each character, i.e., from i=0 to 13, and the character with the smallest Ei is recognized as the read character. Based on the results identified by the identification unit 11 in this way, the device control unit 13
performs sorting and sorting of checks C, totaling of amounts, etc.

The above-mentioned identification unit 11 calculates equation (1) for all characters from stroke 0 to stroke 13, and by checking the position where the maximum peak value appears in the character, it is possible to identify the character. Characters that have no possibility of being identified may be excluded from the calculation of equation (1). In other words, the character candidates to be calculated by formula (1) are selected from the mesh number having the maximum value extracted and held in the maximum value extraction unit 8, thereby shortening the calculation time and increasing the speed. becomes possible. Furthermore, if the minimum Ei is larger than a certain reference value, identification may not be performed and the process may be treated as a reading error.

According to the above embodiment, a digital signal with an amplitude for each mesh of characters is created from the analog signal obtained by reading the characters, and these are compared with the standard pattern of each character. Even if there is a slight error, the effect is relatively small, and the possibility of incorrect character identification is low. The amplitude data for each mesh is based on the maximum peak value within that mesh, so even if the peak position shifts slightly within the mesh, a low peak value will not occur due to noise etc. However, Metsuyu's data is accurate and free of fluctuations. Since the character data is normalized and then compared with the standard pattern data, the comparison is easy to perform and the standard pattern data is easy to create. Since the peak value of each mesh is normalized by dividing it by the maximum peak value, the circuit configuration or program for normalization is easy and high-speed processing is possible.

Note that the above-described circuit configuration can be realized by hardware, software, or a combination thereof.

As described above, according to the present invention, the peak value of each mesh in the horizontal direction of a character is detected and used as character data, which is normalized and then compared with standard pattern data. Character recognition errors are unlikely to occur even if slight fluctuations occur or noise is mixed in. Therefore, MICR
This makes it possible to accurately recognize characters without being affected by differences in printing quality of characters, differences in relative positions of magnetic heads, or changes over time of checks or devices.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of implementing the present invention, and FIG. 2 is a time chart showing a signal when a character with stroke 8 is read. 2...Reading head, 4...Peak detector, 5...
A/D converter, 6... Character cutting section, 7... Memory section, 8... Maximum value extraction section, 9... Normalization control section, 10
...Dividing section, 11... Identification section, 12... Pattern section.

Claims (1)

[Scope of Claims] 1. A method of reading characters written with magnetic ink using a reading head that moves relative to the reading head to obtain an analog signal, and processing this analog signal to recognize the characters, the method comprising: A/D converting the analog signal at each peak position, detecting the peak value for each mesh in the horizontal direction of the character to obtain character data, and normalizing the character data,
A magnetic ink character recognition method characterized by identifying characters by comparing each character with standard pattern data. 2. The magnetic ink character according to claim 1, wherein the character data is normalized by dividing the peak value of each mesh of the character data by the peak value of the maximum absolute value in the character data. Recognition method.